Here is what an LLM that knows nothing after 1930 thinks our world looks like in 2026

Researchers built Talkie, a 13B-parameter model trained only on text published before 1931.

It answers future questions with early-20th-century assumptions, imagining a 2026 shaped by steamships and railroads and underestimating World War II.

The model performs poorly on post-1930 knowledge, showing how a restricted training era changes both worldview and factual prediction.

The work also highlights a practical challenge: building historically bounded AI requires careful filtering and OCR cleanup to avoid later-era contamination.